Core SDK: remove dead/unreferenced code (superseded mixin, empty stubs, dead imports)

[MervinPraison]

Summary

Several files and symbols in praisonaiagents are genuinely unreferenced — a superseded mixin, empty placeholder modules, and dead imports/variables. Removing them is a pure clean-up: no runtime behaviour changes and no public API is touched. This reduces navigation confusion (e.g. a maintainer chasing a "draft" __init__new.py or an orphaned mixin to understand intent).

Current behaviour

All of the following were verified unreferenced across the core package and tests/:

1. praisonaiagents/agent/unified_chat_mixin.py — 151-line UnifiedChatMixin class. It was superseded in place by _execute_unified_chat_completion / _execute_unified_achat_completion in agent/chat_mixin.py (around lines 1353–1543), which is what Agent actually uses. The only references to UnifiedChatMixin are its own definition plus two docstring mentions in chat_mixin.py:1370 and chat_mixin.py:1494. It is not in agent/__init__.py.__all__, not imported by agent.py, and not referenced by any test.
2. Four empty (0-byte) placeholder modules: praisonaiagents/__init__new.py, praisonaiagents/llm/__init__new.py, praisonaiagents/planning/__init__new.py, praisonaiagents/ui/__init__new.py. No code anywhere references the string __init__new.
3. Dead import in praisonaiagents/memory/workflows.py:30 — StepResult is imported from ..workflows.results but immediately shadowed by a local @dataclass class StepResult at line 43 (an intentionally different, incompatible schema used throughout the file). The imported StepResult is never used; the other names on that import line (StepError, WorkflowResult, StepStatus, ErrorStrategy) are used and must stay.
4. Dead module-level variables in praisonaiagents/tools/__init__.py:7-8 — _tools_lock = threading.Lock() and _tools_lazy_cache = {} are declared but never read or written. The actual lazy machinery uses importlib.import_module in __getattr__ and a separate _loaded_classes cache. _tools_lock is never acquired.

Why it matters

Maintenance/clarity: ~150+ lines of orphaned code and four empty modules invite "is this a draft replacement?" confusion, and the dead StepResult import / unused lock imply contracts that do not exist. No import-time or hot-path cost beyond a trivial lock+dict allocation per tools import.

Category

Dead code

Capability preserved

• No user-facing behaviour changes whatsoever. The live execution path (chat_mixin._execute_unified_chat_completion/_achat_completion), the local StepResult dataclass in memory/workflows.py, and the real tools lazy-loader (__getattr__ + _loaded_classes) all remain exactly as-is.

Proposed approach

Remove dead code only:

• Delete agent/unified_chat_mixin.py (and tidy the two stale docstring mentions in chat_mixin.py).
• Delete the four empty __init__new.py files.
• Drop StepResult from the import on memory/workflows.py:30 (keep the local class).
• Delete tools/__init__.py:7-8 (_tools_lock, _tools_lazy_cache).

Resolution sketch

# memory/workflows.py — before
from ..workflows.results import StepResult, StepError, WorkflowResult, StepStatus, ErrorStrategy
# after (StepResult is shadowed by the local dataclass below)
from ..workflows.results import StepError, WorkflowResult, StepStatus, ErrorStrategy

# tools/__init__.py — before
_tools_lock = threading.Lock()
_tools_lazy_cache = {}
# after: removed (unused; real cache is _loaded_classes in __getattr__)

Severity

Low

Validation

Each item confirmed unreferenced via repo-wide grep across praisonaiagents/ and tests/. unified_chat_mixin.py has no __all__ entry, no import, no test reference. The four __init__new.py files are 0 bytes. StepResult from the import is shadowed before any use. _tools_lock/_tools_lazy_cache have zero read/write sites. No public API or robustness behaviour is involved.

Keep unchanged

• chat_mixin._execute_unified_chat_completion / _execute_unified_achat_completion (the live path).
• The local StepResult dataclass in memory/workflows.py and the other imported result types.
• The tools __getattr__ lazy-loader and _loaded_classes cache.
• All protocols, adapters, and extension points.

[praisonai-triage-agent]

Claude finished @MervinPraison's task —— View job • claude/issue-1952-20260613-1157 • [Create PR ➔](https://github.com/MervinPraison/PraisonAI/compare/main...claude/issue-1952-20260613-1157?quick_pull=1&title=Issue%20%231952%3A%20Changes%20from%20Claude&body=This%20PR%20addresses%20issue%20%231952%0A%0AGenerated%20with%20%5BClaude%20Code%5D(https%3A%2F%2Fclaude.ai%2Fcode)

---

✅ Dead Code Successfully Removed from PraisonAI Core SDK

I've completed the cleanup of dead/unreferenced code from the core SDK as described in issue #1952.

Todo List:

• Set up git identity and authentication
• Read AGENTS.md for architecture guidelines
• Verify the dead code claims by examining files
• Remove praisonaiagents/agent/unified_chat_mixin.py
• Remove four empty __init__new.py files
• Remove dead StepResult import from memory/workflows.py
• Remove unused variables from tools/__init__.py
• Clean up stale docstring references
• Run tests to verify no regressions
• Commit and push changes
• Create pull request

✅ Changes Completed

1. Deleted superseded mixin (agent/unified_chat_mixin.py - 151 lines)

   • Functionality superseded by _execute_unified_chat_completion / _execute_unified_achat_completion
   • Cleaned up stale docstring references in chat_mixin.py

2. Deleted 4 empty placeholder files (0 bytes each):

   • praisonaiagents/__init__new.py
   • praisonaiagents/llm/__init__new.py
   • praisonaiagents/planning/__init__new.py
   • praisonaiagents/ui/__init__new.py

3. Removed dead import (memory/workflows.py:30)

   • Removed unused StepResult import that was shadowed by local class

4. Removed unused variables (tools/__init__.py:7-8)

   • Removed _tools_lock and _tools_lazy_cache (never used)
   • Removed unused import threading

🎯 Impact

• No runtime behavior changes - all live execution paths preserved
• No public API changes - no exports modified
• Reduced codebase size by ~160 lines of dead code
• Improved maintainability by removing navigation confusion

Pull Request Created: #1958

---

View Job Run | Branch: claude/issue-1952-20260613-1157

[MervinPraison]

During a follow-up health review of the core SDK, I verified four additional genuinely-unreferenced symbols that fit this issue's "remove dead/unreferenced code" scope (the four empty __init__new.py stubs listed above are confirmed and not repeated here). Each was checked across praisonaiagents/ and tests/; all are pure clean-ups with no runtime behaviour change and no public API impact.

1. praisonaiagents/context/manager.py:97-155 — module-level function deduplicate_topics() (~59 lines, Jaccard-similarity topic de-duplication). Not in any __all__, not imported anywhere, no test references. grep -rn "deduplicate_topics" returns only the definition line.

2. praisonaiagents/llm/unified_adapters.py:144-158 — LiteLLMAdapter._convert_messages_to_prompt() (~15 lines). Never called internally and no test references. (Note: the separate process_stream_response client method tested in tests/unit/test_streaming_integration.py is a different symbol — see item 4 below for the related dead wrapper.)

3. praisonaiagents/agent/memory_mixin.py:13-37 — three module-level fallback helpers _get_console(), _get_live(), _get_display_functions() (~24 lines). They appear to have been copy-pasted from the chat/execution mixins; MemoryMixin has no display/console responsibilities. They are never called inside memory_mixin.py and no other module imports them (only the MemoryMixin class itself is imported by agent.py / agent/__init__.py).

4. praisonaiagents/agent/chat_mixin.py:589-605 — wrapper method _process_stream_response() (~17 lines). grep -rn "_process_stream_response" across the repo finds only this definition; the only test hit (tests/unit/test_streaming_integration.py:40) inspects OpenAIClient.process_stream_response (the underscore-less client method), not this wrapper. The wrapper is never invoked.

Keep unchanged in all four cases: no public API symbol is touched, no behaviour changes, and the surrounding classes/modules (ContextManager, LiteLLMAdapter, MemoryMixin, the chat-streaming path that calls self._openai_client.process_stream_response(...) directly) remain exactly as-is. Happy to fold these into the same clean-up PR.

---

Generated by Claude Code